In the 2002 film “Minority Report,” set in 2054, a sophisticated policing system called PreCrime predicts crimes before they occur. A similar technology is now emerging from South Korea, where researchers have developed an advanced system that combines surveillance cameras (CCTV) and artificial intelligence (AI) to forecast crime risks. This technology is expected to be commercialized by the end of next year.
On July. 18, the Electronics and Telecommunications Research Institute (ETRI) announced the development of ‘Dejaview,’ a system that analyzes CCTV footage and crime statistics to identify potential criminal activity and predict the likelihood of crimes occurring. Dejaview operates on the principle that most crimes follow patterns similar to those of past incidents. By using AI to compare and analyze current situations against historical crime data—including type, method, location, and time—Dejaview assesses the risk of crime.
Dejaview focuses on predicting crimes based on time and location. For example, if a man is seen following a woman late at night, traditional CCTV would simply capture the two individuals walking separately. However, Dejaview evaluates the frequency of similar past crimes in that area and at that time. If the man is observed following the woman across multiple surveillance cameras, the system flags a high risk of crime and alerts the police, who can then respond accordingly. “AI, trained on statistical data from past crimes, automatically analyzes real-time CCTV footage to assess the similarity to known crime scenarios,” explained Kim Geon-woo, a senior researcher in the AI Convergence Security Research Section at ETRI’s Cyber Security Research Division.
To develop Dejaview, ETRI analyzed big data from 32,656 CCTV-recorded incidents and accidents in Seocho-gu, Seoul, over three years starting from 2018. This data was used to create a Crime Prediction Map (PCM), which categorizes crimes by time, location, and type, highlighting areas with high crime potential and displaying real-time crime risk. The PCM boasts an 82.8% accuracy in crime prediction. Additionally, it can forecast the risk of recidivism for individuals under high-risk electronic monitoring (anklets). “We plan to offer this service in collaboration with local governments and control agencies,” ETRI stated, “and aim to commercialize it by the end of next year.”
